List of supported models in cif_direct
models_cif_direct.RdSupported models for the outcome_model argument when using method="direct" in the adjustedcif function.
Details
The following models are directly supported in the outcome_model in the cif_direct function. The first letter in parentheses after the object name is a group indicator. Below the list there are more information for each group.
CSC[A, Required Packages: riskRegression]FGR[B, Required Packages: riskRegression]riskRegression[B, Required Packages: riskRegression]prodlim[B, Required Packages: prodlim, riskRegression]rfsrc[B, Required Packages: randomForestSRC, riskRegression]ARR[B, Required Packages: riskRegression]fit_hal[B, Required Packages: hal9001, riskRegression]fastCrr[C, Required Packages: fastcmprsk]comp.risk[C, Required Packages: timereg]cifreg[D, Required Packages: mets]Any model with a fitting S3 prediction method or a valid
predict_funcan be used as well. See below.
Group A: The direct adjusted cumulative incidences are estimated directly using the ate function. Additional arguments supplied using the ... syntax are passed to the ate function.
Group B: The predictRisk function is used to obtain predicted cumulative incidences, which are then used in the G-Computation step. Additional arguments supplied using the ... syntax are passed to the predictRisk function.
Group C: Custom code is used to do the estimation. Additional arguments supplied using the ... syntax are currently not supported.
Group D: Models fit using cifreg or cifregFG from the mets package. The direct adjusted cumulative incidences are estimated using the survivalG function of the mets package, which also provides influence function based standard errors, so that conf_int=TRUE can be used without bootstrapping. Since these standard errors are calculated separately for each point in time, it may be useful to specify a smaller number of points in time using the times argument in this case. Bootstrapping and data with events at time 0 are currently not supported for these models. Additional arguments supplied using the ... syntax are not used.
It is sometimes possible to use models even if they are not listed here. There are two ways to make this work. The first one is to use the models S3 predict method. This works if the predict function contains the arguments object, newdata, times and cause and returns a matrix of predicted cause-specific cumulative incidences. The matrix should be of size nrow(data) * length(times), where each row corresponds to a row in the original dataset and each column to one point in time. The matrix should contain the cause-specific cumulative incidences predicted by the model given covariates. If no such predict method exists the only option left is to write your own function which produces the output described above and supply this function to the predict_fun argument.
If you think that some important models are missing from this list, please file an issue on the official github page with a specific feature request (URL can be found in the DESCRIPTION file) or contact the package maintainer directly using the given e-mail address.